AI in cybersecurity is exposing shadow AI governance gaps

At a glance

What this is: This podcast discussion argues that AI in cybersecurity is creating a shadow AI governance gap, where unseen AI features and agentic identities can act without clear inventory, ownership, or oversight.

Why it matters: IAM and security teams now have to govern AI-enabled access and decisions with the same discipline they apply to human and machine identities, or they will lose control of provenance, accountability, and access scope.

Context

AI in cybersecurity is no longer only a tooling question. The governance gap emerges when AI features, autonomous assistants, and other embedded capabilities gain access paths that security teams cannot clearly inventory, own, or review.

The article’s central concern is shadow AI, meaning AI-driven capability that is present in the environment but not properly governed. For identity programmes, that shifts the problem from static account control to ongoing visibility, provenance, and access accountability across people, machines, and AI-enabled systems.

Key questions

Q: What breaks when shadow AI is not included in identity governance?

A: When shadow AI is excluded, the organisation loses discovery, ownership, and enforcement at the same time. Unmanaged local agents can access cloud and SaaS resources without being enrolled in policy, which means no one can attest to their privileges or revoke them cleanly. The first failure is visibility, and the second is accountability.

Q: Why do AI agents require stronger identity controls than standard applications?

A: AI agents can choose actions, call tools, and chain operations, so their identity is not just a login mechanism. If they are overprivileged, one prompt injection or workflow abuse can turn into broad enterprise misuse. Teams should therefore constrain agent permissions, use short-lived credentials, and treat agent access as privileged by default.

Q: How do organisations keep human oversight meaningful in AI workflows?

A: Human oversight stays meaningful only when humans have enough context, time, and authority to intervene. If the AI output is acted on automatically or too quickly to challenge, oversight becomes ceremonial. Effective oversight requires review points, clear escalation rights, and the ability to halt or reverse the decision.

Q: What is the difference between human IAM and agentic AI governance?

A: Human IAM is built around users, sessions, and relatively predictable access patterns. Agentic AI governance must also control memory, delegation, and machine-speed actions that can continue after the original request is made. That means identity policy has to follow the task lifecycle, not just the login event, and must cover both authorization and state handling.

Technical breakdown

Why shadow AI breaks identity inventory models

Shadow AI creates an identity problem because the control point is no longer a named user account or obvious service principal. AI features can be embedded in applications, workflows, or platforms, making them easy to miss in inventories that were designed around known accounts and documented ownership. Once an AI capability can send emails, access data, or trigger actions, it behaves like an identity path even when no one has formally classified it that way. Traditional IGA breaks here because it assumes the organisation already knows what exists and who owns it. Practical implication: inventory AI-enabled access as a first-class identity estate, not as an application feature set.

Practical implication: discover AI-enabled access paths as part of the identity estate, not as optional application metadata.

Agentic identity needs provenance and lineage controls

Agentic AI introduces a different identity pattern from conventional machine access because the system may choose actions at runtime and chain multiple tools together. That creates provenance and lineage requirements: teams need to know which agent initiated an action, which tools it used, what data it touched, and how one decision led to the next. Without that traceability, accountability dissolves into a series of opaque automations. The article is right to treat agentic AI as a new identity category because the governance burden is no longer just authentication. It becomes continuous attribution of intent, action, and delegation across a dynamic runtime path. Practical implication: design controls around lineage, not just login or token issuance.

Practical implication: enforce lineage and attribution controls for agent actions, not just access authentication.

Human oversight remains the final control boundary

The article’s strongest operational point is that human judgment cannot be mechanised away in security governance. Automation can absorb repetitive work, but oversight is still needed where decisions affect risk acceptance, anomalous behaviour, or access scope changes. That matters because AI-driven systems can amplify mistakes faster than human-paced review cycles can catch them. In governance terms, the issue is not whether humans should disappear from the loop, but where human approval, escalation, or exception handling must remain mandatory. The programme design challenge is to place oversight where the risk is irreversible. Practical implication: reserve human decision rights for access changes and exception handling that AI should not close autonomously.

Practical implication: keep human decision rights on high-impact access changes and exception handling.

NHI Mgmt Group analysis

Shadow AI is an identity governance problem before it is an AI problem. When organisations cannot inventory AI features, they cannot govern their access paths, ownership, or review cycles. That breaks the basic IGA assumption that every active identity-like thing is known, named, and assigned. The implication is that AI discovery becomes a prerequisite for identity control, not an adjacent security task.

Agentic AI should be treated as a distinct identity class because it changes how actions are attributed. A conventional machine identity authenticates and executes. An agent can select tools, chain actions, and change the sequence of execution at runtime, which makes provenance and lineage central to governance. The implication is that access governance must move beyond account states and start tracking runtime delegation.

Human oversight is the control that AI adoption tries to compress but cannot eliminate. The article correctly frames judgment, empathy, communication, and education as capabilities that automation does not replace. For identity programmes, that means review, exception approval, and accountability retain a human endpoint even when the work is machine-assisted. The implication is that governance models should define where human discretion is mandatory, not assume it will scale away.

Identity hygiene is becoming AI hygiene because the attack surface now includes unmanaged intelligence. Shadow AI is not merely a visibility issue. It is a provenance, ownership, and access-scoping issue that can create unauthorised decision paths before teams realise an AI system exists. The implication is that identity security teams must extend lifecycle governance to AI-enabled capabilities with the same rigor they apply to credentials and privileged access.

From our research library:

• Companies are dedicating an average of 32.4% of their security budgets to secrets management and code security, with US organisations leading at 40.8%, according to the State of Secrets in AppSec.
• Read next: Shadow AI and AI Agent Discovery Guide

What this signals

Shadow AI discovery now belongs inside identity operations. If AI features can act without clear inventory, then recertification and offboarding logic will always be behind the real risk. Security teams should treat discovery of AI-enabled access paths as an operational control, not a periodic audit task.

Identity programmes need a provenance layer for agentic behaviour. Runtime delegation, tool selection, and action lineage have to be visible if organisations want to preserve accountability across AI-assisted workflows. That is the point where IAM, IGA, and AI governance converge into one operating model.

For practitioners

• Map AI-enabled access paths Inventory every application, workflow, and platform feature that can call models, agents, or automated decision services, then assign an owner and a control boundary for each one.
• Classify agentic identities separately Create a separate identity class for autonomous agents so provenance, tool use, and delegated authority are governed distinctly from standard service accounts.
• Require lineage logging for AI actions Capture which agent or AI feature initiated an action, which tools it invoked, what data it touched, and whether a human approved the outcome.
• Set human decision boundaries Define the access changes, exceptions, and high-impact actions that must stay under human review even when the surrounding workflow is automated.
• Rework access reviews for dynamic AI use Move review logic from static account recertification toward evidence of AI usage, data access, and delegated actions that can appear and disappear quickly.

Key takeaways

• Shadow AI creates governance gaps because organisations often cannot see or own the AI-enabled access paths already operating inside business tools.
• The article connects AI risk to identity control by showing that provenance, ownership, and continuous oversight matter as much as authentication.
• Security teams need to extend identity governance into discovery, lineage, and human decision boundaries before AI-driven workflows become routine.

Key terms

• Shadow AI: AI agents, copilots, or connected tools operating without full visibility or governance from security teams. Shadow AI becomes an identity problem when those systems authenticate with unmanaged tokens, service accounts, or OAuth apps that can reach production resources.
• Agentic Identity: An agentic identity is a non-human identity used by an autonomous system that can act, call tools, and access data with execution authority. It needs the same governance discipline as other privileged identities, plus runtime context, ownership mapping, and revocation paths.
• Identity Provenance: Identity provenance is the record of how an agent was created, what authority it received, and what actions it performed over time. It turns agent activity into an auditable chain of trust that supports compliance, incident response, and post-event accountability.
• Human Oversight: Human oversight is the requirement that a person remains responsible for reviewing, approving, or correcting AI-driven output before it causes a material action. In governance terms, it is the control that prevents automation from becoming unowned authority.

Deepen your knowledge

NHI governance, agentic AI identity, and machine identity security are core topics in our NHI Foundation Level course, the industry's only accredited NHI security programme. If you are building or maturing an identity security programme, it is worth exploring.